Metrics H andbehavioursas sk1 2on loopgroups · 2017-02-08 ·...
Transcript of Metrics H andbehavioursas sk1 2on loopgroups · 2017-02-08 ·...
Journal of Functional Analysis 213 (2004) 440–465
Metrics Hs and behaviours as sk1=2 onloop groups
Shizan Fang
Laboratoire de Topologie, Universite de Bourgogne, 9 avenu Alain Savary, BP 47870, Dijon Cedex,
France and Department of Mathematics, Beijing Normal University, Beijing 100875, China
Received 10 June 2003; accepted 23 July 2003
Communicated by Paul Malliavin
Abstract
For all metricHs with s41=2; the heat measures on free loop groups have been constructedby P. Malliavin. The geometric stochastic analysis is well developed. The metricH1=2 is critical
and involves naturally in Mathematical Physics. The Ricci tensors Rics play the fundamental
role in these situations. In this note, we shall study the behaviour of Rics as sk1=2:r 2003 Elsevier Inc. All rights reserved.
0. Introduction
Let G be a semi-simple compact Lie group and G its Lie algebra, endowed with an
AdG invariant metric /;SG: Denote by D ¼ � d2
dy2the Laplace operator on the circle
S1: For s40; define on CNðS1;GÞ the following metric:
jhj2s ¼Z
S1
/ð1þ DÞsh; hSG
dy2p
:
Writing hACNðS1;GÞ in Fourier series,
hðyÞ ¼XnAZ
hðnÞeffiffiffiffiffi�1
pny where hðnÞ ¼
ZS1
hðyÞe�ffiffiffiffiffi�1
pny dy2p
; ð0:1Þ
ARTICLE IN PRESS
E-mail address: [email protected].
0022-1236/$ - see front matter r 2003 Elsevier Inc. All rights reserved.
doi:10.1016/j.jfa.2003.07.006
then
jhj2s ¼XnAZ
ð1þ n2ÞsjhðnÞj2G: ð0:2Þ
Let
HsðGÞ ¼ fhAL2ðS1;GÞ; jhjsoþNg: ð0:3Þ
By Sobolev embedding theorem, HsðGÞCCðS1;GÞ for s41=2: In this case, there
is a Gaussian measure ms on CðS1;GÞ such that ðCðS1;GÞ;HsðGÞ; msÞ becomes anabstract Wiener space. Let fun; nX1g be an orthonormal basis of HsðRÞ;fe1;y; edg an orthonormal basis of G: Define
en;aðyÞ ¼ unðyÞea; nX1; a ¼ 1;y; d: ð0:4Þ
Then fen;ag is an orthonormal basis of HsðGÞ: In what follows, we shall work with
this type of basis. Let GðsÞðy1; y2Þ be the Green function on S1 associated to the
differential operator ð1þ DÞs; that is the solution of
ð1þ DÞsGðsÞðy1; Þ ¼ dy1 ; ð0:5Þ
where dy1 is the Dirac mass at y1: Relation (0.5) means that for any uAHsðGÞ;ðGðsÞðy1; Þ; uÞs ¼ uðy1Þ: ð0:6Þ
Therefore for any orthonormal basis fung of HsðRÞ; we haveGðsÞðy1; y2Þ ¼
Xn
unðy1Þunðy2Þ:
If we take
u0 ¼ 1; u2n�1ðyÞ ¼ffiffiffi2
pcos ny
ð1þ n2Þs=2; u2nðyÞ ¼
ffiffiffi2
psin ny
ð1þ n2Þs=2; nX1; ð0:7Þ
we obtain
GðsÞðy1; y2Þ ¼XnAZ
effiffiffiffiffi�1
pnðy1�y2Þ
ð1þ n2Þs : ð0:8Þ
Let fxn;a; nX1; a ¼ 1;y; dg be a sequence of independent standard real-valued
Brownian motion, defined on a probability space ðO;F;PÞ: Considerxðt; yÞ ¼
Xn;a
xn;aðtÞen;aðyÞ: ð0:9Þ
It is well known that for s412the above series converges uniformly with respect to
ðt; yÞA½0; 1� S1: We have for a; bAG;
Eð/xðt; y1Þ; aSG/xðs; y2Þ; bSGÞ ¼ t4sGðy1; y2Þ/a; bSG: ð0:10Þ
ARTICLE IN PRESSS. Fang / Journal of Functional Analysis 213 (2004) 440–465 441
t-xðt; Þ is a Brownian motion on CðS1;GÞ with ð ; Þs as covariance operator. For
yAS1; consider the following s.d.e.
dtgxðt; yÞ ¼ gxðt; yÞ 3 dtxðt; yÞ; gxð0; yÞ ¼ e: ð0:11Þ
It has been proved in [Ma] that ðt; yÞ-gxðt; yÞ has a continuous version that we
denote by the same notation. Then t-gxðt; Þ is a continuous process on LðGÞ :¼CðS1;GÞ: Denote by nðsÞt the law of x-gxðt; Þ on LðGÞ: The geometric stochasticanalysis on nðsÞt is well developed for s41
2: See [Ca,DL,Dr,Fa] for s ¼ 1 and [FG,In]
for s41=2: The case where s ¼ 12is critical but it is involved in Mathematical Physics.
On the based loop group LeðGÞ ¼ LðGÞ=G; there is a complex structure J which iscompatible with the metric H1=2 (see [Fr,PS]). However the H1=2 loops in G do not
form a smooth manifold. It is the limit case of HsðGÞ as sk1=2:The paper is organized as follows. In Section 1, we shall prove that the
expression of Ricci tensor obtained in [In] for s41=2 is valid for sX1=2 and
Ricsðz1; z2Þ converges to Ric1=2ðz1; z2Þ when s-1=2: In Section 2, we shall simplifythe Inahama’s expression and prove the boundedness of Rics for s41=2 by
estimating as well as possible the norm jjRicsjjop: The boundedness of Ric1=2 will be
established in Section 3. In Section 4, we shall discuss the uniform boundednessabout jjRicsjjop:
1. Ricci tensor in HsðGÞ for sX12
Let z1; z2ACNðS1;GÞ: Define the bracket by
½z1; z2�ðyÞ ¼ ½z1ðyÞ; z2ðyÞ�: ð1:1Þ
Following [Fr, p. 230], we define for s40;
rz1z2 ¼1
2f½z1; z2� þ ð1þ DÞ�s½z1; ð1þ DÞs
z2� þ ð1þ DÞ�s½z2; ð1þ DÞsz1�g: ð1:2Þ
We see that rz1z2ACNðS1;GÞ and satisfies
ðrz1z2; hÞs ¼1
2fð½z1; z2�; hÞs � ð½z1; h�; z2Þs � ð½z2; h�; z1Þsg: ð1:3Þ
Proposition 1.1 (Inahama [In]). Let z1; z2ACNðS1;GÞ; then
ðrz1z2ÞðyÞ ¼XkAZ
Xpþq¼k
½z1ðpÞ; z2ðqÞ�Tp;q
!effiffiffiffiffi�1
pky; ð1:4Þ
ARTICLE IN PRESSS. Fang / Journal of Functional Analysis 213 (2004) 440–465442
where
Tp;q ¼ 1
2
ð1þ ðp þ qÞ2Þs � ð1þ p2Þs þ ð1þ q2Þs
ð1þ ðp þ qÞ2Þs: ð1:5Þ
Proof. We have
½z1; z2� ¼XkAZ
Xpþq¼k
½z1ðpÞ; z2ðqÞ� !
effiffiffiffiffi�1
pky;
ð1þ DÞ�1½z1; ð1þ DÞsz2� ¼
XkAZ
Xpþq¼k
½z1ðpÞ; z2ðqÞ�ð1þ q2Þs
ð1þ ðp þ qÞ2Þs
!effiffiffiffiffi�1
pky;
ð1þ DÞ�1½z2; ð1þ DÞsz1� ¼ �
XkAZ
Xpþq¼k
½z1ðpÞ; z2ðqÞ�ð1þ p2Þs
ð1þ ðp þ qÞ2Þs
!effiffiffiffiffi�1
pky:
Therefore by (1.2), we obtain the result (1.4). &
We shall compute the associated Ricci curvature. For z1; z2; z3ACNðS1;GÞ; define
Rðz1; z2Þz3 ¼ rz1rz2z3 �rz2rz1z3 �r½z1;z2�z3: ð1:6Þ
By Freed [Fr], the two-step trace of ðh1; h2Þ-ðRðz1; h1Þh2; z2Þs does exist if s414:
More precisely, the series
Ricsðz1; z2Þ :¼XnX0
Xd
a¼1ðRðz1; en;aÞen;a; z2Þs
exists for s41=4: For a; bAG; let
Kða; bÞ ¼ TraceðadðaÞ 3 adðbÞÞ:
When G is semi-simple, �K is an AdG invariant metric on G: For simplicity ofnotations, we shall take �K as the AdG invariant metric on G in the sequel.
Lemma 1.2. Let
AðsÞq ðcÞ ¼ �Tc;�qTq;c�q þ Tcþq;�q
ð1þ q2Þs for sX1=2: ð1:7Þ
ARTICLE IN PRESSS. Fang / Journal of Functional Analysis 213 (2004) 440–465 443
Then there exists a constant C40 (independent of c) such that
sup1=2psp1
jAðsÞq ðcÞjpC
c2
q2for jqjX3c: ð1:8Þ
Proof. Set
NðsÞq ðcÞ ¼ ð1þ c2Þ2s þ ð1þ q2Þ2s � ð1þ ðc� qÞ2Þ2s � 2ð1þ c2Þsð1þ q2Þs
þ 2ð1þ c2Þsð1þ ðc� qÞ2Þs � 2ð1þ ðcþ qÞ2Þsð1þ ðc� qÞ2Þs
þ 2ð1þ q2Þsð1þ ðc� qÞ2Þs;
and
DðsÞq ðcÞ ¼ ð1þ q2Þsð1þ c2Þsð1þ ðc� qÞ2Þs:
Then by (1.5) and (1.7), we get AðsÞq ðcÞ ¼ 1
4
NðsÞq ðcÞ
DðsÞq ðcÞ
: Remark that there exists a constant
C40 (independent of gA½�2; 4�Þ such that
jð1þ xÞg � ð1þ gxÞjpCx2; jxjp12; gA½�2; 4�;
which will be written in the form
ð1þ xÞg ¼ 1þ gx þ Oðx2Þ; jxjp1
2; ð1:9Þ
where O is uniform with respect to gA½�2; 4�: Let c40 and jqjX3c: To fix the idea,consider q40: Then we have
ð1þ q2Þ2s ¼ q4s 1þ O1
q2
� �� �:
To estimate the term ð1þ ðq � cÞ2Þs; remark that for qX3c; q � cX2cX2: Applying(1.9), we get
1þ 1
ðq � cÞ2
!2s
¼ 1þ 2s
ðq � cÞ2þ O
1
ðq � cÞ2
!
¼ 1þ O1
q2
� �; as q � cXq=2;
and
ðq � cÞ4s ¼ q4s 1� 4sc
qþ O
c2
q2
� �� �:
ARTICLE IN PRESSS. Fang / Journal of Functional Analysis 213 (2004) 440–465444
Therefore
ð1þ ðq � cÞ2Þ2s ¼ q4s 1� 4sc
qþ O
c2
q2
� �� �:
In the same way, we find
ð1þ ðq þ cÞ2Þs ¼ q2s 1þ 2sc
qþ O
c2
q2
� �� �;
ð1þ ðq � cÞ2Þsð1þ ðq þ cÞ2Þs ¼ q4s 1þ Oc2
q2
� �� �;
ð1þ q2Þsð1þ ðc� qÞ2Þs ¼ q4s 1� 2sc
qþ O
c2
q2
� �� �:
A straightforward calculation yields
NsqðcÞ ¼ ð1þ c2Þ2s � 4scð1þ c2Þs
q2s�1 þ c2Oðq4s�2Þ:
For term DðsÞq ðcÞ; we have for qX3c; ð1þ ðq � cÞ2Þs
Xðq=2Þ2s: Then
DðsÞq ðcÞXð1þ c2Þs
q4s 1
2
� �2s
:
Therefore there exists a constant C40 such that
sup1=2psp1
NsqðcÞ
DsqðcÞ
����������pC
c2
q2for qX3c: ð1:10Þ
For qp� 3c; changing q into �q; the term ðq � cÞ2 goes into ðq þ cÞ2 and ðq þ cÞ2
into ðq � cÞ2: So estimate (1.10) remains true for qp� 3c: The proof of (1.8) iscomplete. &
Lemma 1.3. There exists a constant C40 such that
sup1=2psp1
XqAZ
jAðsÞq ðcÞj
!pCc2: ð1:11Þ
Proof. Remark that for jqjp3c;
jNsqðcÞjpCð1þ c2Þ2s for some constant C40:
ARTICLE IN PRESSS. Fang / Journal of Functional Analysis 213 (2004) 440–465 445
Then for sX1=2;
Xjqjo3c
NsqðcÞ
DsqðcÞ
����������pCð1þ cÞs
Xjqjo3c
1ffiffiffiffiffiffiffiffiffiffiffiffiffi1þ q2
p ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi1þ ðq � cÞ2
q :
We have
Xcq¼0
1ffiffiffiffiffiffiffiffiffiffiffiffiffi1þ q2
p ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi1þ ðq � cÞ2
q p2X½c=2�þ1q¼0
1ffiffiffiffiffiffiffiffiffiffiffiffiffi1þ q2
p ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi1þ ðq � cÞ2
q p2ð½c=2� þ 2Þ½c=2� � 1
p4;
where ½c=2� denotes the integral part of c=2; and
X3cq¼cþ1
1ffiffiffiffiffiffiffiffiffiffiffiffiffi1þ q2
p ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi1þ ðq � cÞ2
q p2:
It follows that there exists a constant C140 such that
Xjqjo3c
NsqðcÞ
DsqðcÞ
����������pC1ð1þ c2Þs:
According to (1.8), we obtain (1.11). &
Theorem 1.4 (see Inahama [In]). Let z1; z2ACNðS1;GÞ: For sX1=2; we have
Ricsðz1; z2Þ ¼XcAZ
ð1þ c2ÞsXqAZ
AðsÞq ðcÞ
!/z1ðcÞ; %z2ðcÞS; ð1:12Þ
where ‘‘bar’’ denotes the complex conjugation and the inner product of G is extended by
complex linearity to its complexification GC:
Proof. The expression (1.12) comes from [In, (2.31)] with an opposite sign. In whatfollows, we shall give a proof valid for all case sX1=2: By (1.4), for zAHsðRÞ#G; wehave: rzz ¼ 0: Then Rðz1; zÞz ¼ �rzrz1z �r½z1;z�z: We have
rzrz1z ¼XcAZ
Xrþk¼c
Xpþq¼k
½zðrÞ; ½z1ðpÞ; zðqÞ��Tp;qTr;k
!effiffiffiffiffi�1
pcy;
r½z1;z�z ¼XcAZ
Xkþr¼c
Xpþq¼k
½½z1ðpÞ; zðqÞ�; zðrÞ�Tk;r
!effiffiffiffiffi�1
pcy:
ARTICLE IN PRESSS. Fang / Journal of Functional Analysis 213 (2004) 440–465446
Then
Rðz1; zÞz ¼XcAZ
Xrþk¼c
Xpþq¼k
½zðrÞ; ½z1ðpÞ; zðqÞ��ð�Tp;qTr;k þ Tk;rÞ !
effiffiffiffiffi�1
pcy:
Under the hypothesis of z1; zACNðS1;GÞ; the above multiple series converge
absolutely. Let z2ACNðS1;GÞ: Consider en;a defined in (0.4) and (0.7), then
ðRðz1; en;aÞen;a; z2Þs
¼XcAZ
ð1þ c2ÞsX
pþqþr¼c
/½ea; ½z1ðpÞ; ea��; %z2ðcÞSunðrÞunðqÞð�Tp;qTr;pþq þ Tpþq;rÞ !
:
Therefore
Xd
a¼1ðRðz1; en;aÞen;a; z2Þs
¼Xc
ð1þ c2ÞsX
pþqþr¼c
/z1ðpÞ; %z2ðcÞSunðrÞunðqÞð�Tp;qTr;pþq þ Tpþq;rÞ:
Remark that
u2n�1ðrÞu2n�1ðqÞ þ u2nðrÞu2nðqÞ
is equal to 1ð1þn2Þs if r ¼ �q ¼ n or r ¼ �q ¼ �n; and equal to zero otherwise.
Therefore
X2N
n¼0
Xd
a¼1ðRðz1; en;aÞen;a; z2Þs
¼XN
n¼�N
Xc
ð1þ c2Þs
ð1þ n2Þs ðTc;�nTn;c�n � Tc�n;nÞ/z1ðcÞ; %z2ðcÞS:
Splitting the sum into two terms and changing the index n into �n in the secondterm, the above sum goes to
XN
n¼�N
Xc
ð1þ c2Þs
ð1þ n2Þs ðTc;�nTn;c�n � Tcþn;�nÞ/z1ðcÞ; %z2ðcÞS:
Therefore
X2N
n¼0
Xd
a¼1ðRðz1; en;aÞen;a; z2Þs
¼Xc
ð1þ c2ÞsXN
n¼�N
AðsÞn ðcÞ
!/z1ðcÞ; %z2ðcÞS: ð1:13Þ
ARTICLE IN PRESSS. Fang / Journal of Functional Analysis 213 (2004) 440–465 447
Now by (1.11), for z1; z2ACNðS1;GÞ;XcAZ
ð1þ c2Þ2X
q
jAðsÞq ðcÞj
!j/z1ðcÞ; %z2ðcÞSjoþN:
Letting N-þN in (1.13), we get the result (1.12) by Lebesgue dominatedconvergence theorem. &
Theorem 1.5. For z1; z2ACNðS1;GÞ;
limsk1=2
Ricsðz1; z2Þ ¼ Ric1=2ðz1; z2Þ: ð1:14Þ
Proof. By (1.8), we see that the seriesP
q AðsÞq ðcÞ converges uniformly in sA½1
2; 1�:
Then
limsk1=2
Xq
AðsÞq ðcÞ ¼
Xq
Að1=2Þq ðcÞ:
Now for z1; z2ACNðS1;GÞ;XcAZ
ð1þ c2Þ2j/z1ðcÞ; %z2ðcÞSjoþN:
Again according to (1.11), we get
limsk1=2
Ricsðz1; z2Þ ¼XcAZ
ð1þ c2Þ1=2XqAZ
Að1=2ÞðcÞ !
/z1ðcÞ; %z2ðcÞS
¼Ric1=2ðz1; z2Þ: &
2. Boundedness of Rics for s41=2
In [In], it has been proved that for s41=2; there exists a constant Cs40 such that
jRicsðz1; z2ÞjpCsjz1jsjz2js; z1; z2AHsðGÞ: ð2:1Þ
It follows that there exists a bounded operator Rics :HsðGÞ-HsðGÞ such that
ðRicsz1; z2Þs ¼ Ricsðz1; z2Þ:
In what follows, we shall simplify expression (1.12) and estimate as well as possiblethe norm jjRicsjjop:
ARTICLE IN PRESSS. Fang / Journal of Functional Analysis 213 (2004) 440–465448
Proposition 2.1. Define
BðsÞq ðcÞ ¼ ð1þ c2Þs
ð1þ ðc� qÞ2Þsð1þ q2Þs; ð2:2Þ
DðsÞq ðcÞ ¼ ½ð1þ q2Þs � ð1þ ðcþ qÞ2Þs� � ½ð1þ ðq � cÞ2Þs � ð1þ qÞ2Þs�
ð1þ c2Þsð1þ q2Þs : ð2:3Þ
Then for s41=2;
4XqAZ
AðsÞq ðcÞ ¼
XqAZ
BðsÞq ðcÞ þ
XqAZ
DðsÞq ðcÞ: ð2:4Þ
Proof. The term NðsÞq ðcÞ defined in proof of Lemma 1.2 can be written in
the form:
NðsÞq ðcÞ ¼ 2ð1þ c2Þsð1þ ðc� qÞ2Þs � 2ð1þ c2Þsð1þ q2Þs
þ ð1þ c2Þ2s þ ½ð1þ q2Þ2s � ð1þ ðcþ qÞ2Þsð1þ ðc� qÞ2Þs�
þ ½2ð1þ q2Þs � ð1þ ðcþ qÞ2Þs � ð1þ ðc� qÞ2Þs�ð1þ ðc� qÞ2Þs:
Then
4AðsÞq ðcÞ ¼ 2
ð1þ q2Þs �2
ð1þ ðq � cÞ2Þsþ ð1þ c2Þs
ð1þ q2Þsð1þ ðq � cÞ2Þs
þ ð1þ q2Þ2s � ð1þ ðcþ qÞ2Þsð1þ ðc� qÞ2Þs
ð1þ c2Þsð1þ q2Þsð1þ ðq � cÞ2Þs
þ 2ð1þ q2Þs � ð1þ ðcþ qÞ2Þs � ð1þ ðc� qÞ2Þs
ð1þ c2Þsð1þ q2Þs :
For s41=2;P
qAZ1
ð1þq2Þs ¼P
qAZ1
ð1þðq�cÞ2Þs: Therefore in order to get (2.4), it is
sufficient to prove that
XqAZ
ð1þ q2Þ2s � ð1þ ðcþ qÞ2Þsð1þ ðc� qÞ2Þs
ð1þ c2Þsð1þ q2Þsð1þ ðq � cÞ2Þs¼ 0: ð2:5Þ
ARTICLE IN PRESSS. Fang / Journal of Functional Analysis 213 (2004) 440–465 449
Remark firstly that the series in (2.5) converges absolutely if s41=2: In fact for
0osp1; using the inequality jxs � ysjpjx � yjs; we have
jð1þ q2Þ2s � ð1þ ðcþ qÞ2Þsð1þ ðc� qÞ2Þsj
pj2c2 þ c4 � 2c2q2js
Bð2c2Þsq2s; jqj-þN:
Now
Xkcþc
q¼kcþ1
ð1þ q2Þs
ð1þ ðq � cÞ2Þs� ð1þ ðcþ qÞ2Þs
ð1þ q2Þs
!
¼Xkc
q¼kc�cþ1
ð1þ ðcþ qÞ2Þs
ð1þ q2Þs �Xkcþc
q¼kcþ1
ð1þ ðcþ qÞ2Þs
ð1þ q2Þs :
Then
XN
k¼�N
Xkcþc
q¼kcþ1
ð1þ q2Þs
ð1þ ðq � cÞ2Þs� ð1þ ðcþ qÞ2Þs
ð1þ q2Þs
!" #
¼X�Nc
q¼�Nc�cþ1
ð1þ ðcþ qÞ2Þs
ð1þ q2Þs �XNcþc
q¼Ncþ1
ð1þ ðcþ qÞ2Þs
ð1þ q2Þs
which converges to c� c ¼ 0 as N-þN: We get therefore (2.5). &
Lemma 2.2. Define for s41=2;
as ¼XqAZ
1
ð1þ q2Þs: ð2:6Þ
Then we have the following estimate
supcAZ
XqAZ
ð1þ c2Þs
ð1þ ðc� qÞ2Þsð1þ q2Þspð1þ 4sÞas: ð2:7Þ
Proof. Let c40: Then
Xqo0
ð1þ c2Þs
ð1þ ðc� qÞ2Þsð1þ q2ÞspXqo0
1
ð1þ q2Þs;
XqXc
ð1þ c2Þs
ð1þ ðc� qÞ2Þsð1þ q2ÞspXqXc
1
ð1þ ðc� qÞ2Þs¼XqX0
1
ð1þ q2Þs:
ARTICLE IN PRESSS. Fang / Journal of Functional Analysis 213 (2004) 440–465450
We have
Xc�1q¼0
ð1þ c2Þs
ð1þ ðc� qÞ2Þsð1þ q2Þs
¼X½c=2�q¼0
ð1þ c2Þs
ð1þ ðc� qÞ2Þsð1þ q2Þsþ
Xc�1q¼½c=2�þ1
ð1þ c2Þs
ð1þ ðc� qÞ2Þsð1þ q2Þs:
It is easy to see that the first term on the right side is dominated by 22sP
qX01
ð1þq2Þs;
while the second term is dominated by 22sP
q401
ð1þq2Þs: Now combining these
estimates, we obtain (2.7). &
Lemma 2.3. Let 12osp1: We haveX
qAZ
jDðsÞq ðcÞjpð2þ 4sÞas: ð2:8Þ
Proof. Let c40: Remark that for 12osp1; the function f ðxÞ ¼ ð1þ xÞs is concave
and gðxÞ ¼ ð1þ x2Þs is convex. We have
2ð1þ c2Þspð1þ ðcþ qÞ2Þs þ ð1þ ðc� qÞ2Þsp2ð1þ c2 þ q2Þs: ð2:9Þ
Therefore for 0pqpc;
jð1þ ðcþ qÞ2Þs þ ð1þ ðc� qÞ2Þs � 2ð1þ q2Þsj
p2½ð1þ c2 þ q2Þs � ð1þ q2Þs�p2cs:
It follows that
Xcq¼0
jDðsÞq ðcÞjpas:
For q4c; remark that g00ðxÞ ¼ 2sð1þ x2Þs�2 þ 2sð2s � 1Þx2ð1þ x2Þs�1: Write down
ð1þ ðq þ cÞ2Þs � ð1þ q2Þs ¼ 2sg1ð1þ g21Þsc; g1A�q; q þ c½;
ð1þ q2Þs � ð1þ ðq � cÞ2Þs ¼ 2sg2ð1þ g22Þsc; g2A�q � c; q½:
Then there exists bA�g1; g2½C�q � c; q þ c½ such that
j½ð1þ ðq þ cÞ2Þs � ð1þ q2Þs� � ½ð1þ q2Þs � ð1þ ðq � cÞ2Þs�j
p4sc2ð1þ b2Þs�2 þ 4sð2s � 1Þc2ð1þ b2Þs�1 ð2:10Þ
ARTICLE IN PRESSS. Fang / Journal of Functional Analysis 213 (2004) 440–465 451
which is smaller than 8sc2
ð1þðq�cÞ2Þ1�s: It follows that
Xq4c
jDðsÞq ðcÞjp 8s
Xq4c
c2
ð1þ c2Þsð1þ q2Þsð1þ ðq � cÞ2Þ1�s
¼ 8sXq40
c2
ð1þ c2Þsð1þ ðq þ cÞ2Þsð1þ q2Þ1�s:
Remarking that ð1þ ðq þ cÞ2ÞsXð1þ c2Þ1�sð1þ q2Þ2s�1; we get
Xq4c
jDðsÞq ðcÞjp8s
Xq40
1
ð1þ q2Þsp4sas: &
Proposition 2.4. For s41=2; we have
jjRicsjjoppð3þ 4s þ 4sÞas
4: ð2:11Þ
Proof. Using expression (1.12) and estimates (2.4), (2.7) and (2.8), we get
jðRicsz1; z2Þsjpð3þ 4s þ 4sÞas
4jz1jsjz2js: &
It has been proved in [In] (see also [FG]) that z-rzh is Hilbert–Schmidt operatorin HsðGÞ for s41=2 and jjrhjjHSpCsjhjs: This is the key point to develop the
geometric analysis on LðGÞ (see [Dr,Fa]). In this case, there exists a boundedoperator K :HsðGÞ-HsðGÞ such that
ðKz1; z2Þs ¼ �Xn;m
Xd
a;b¼1ðren;az1; em;bÞsðrem;bz2; en;aÞs: ð2:12Þ
The operator K involved naturally in [Fa]. For H1 metric, it is known that K is
identical to Ric1 by their explicit expression (see [FF]). In what follows, we shallprove that it is also true for s41=2:
Proposition 2.6. K ¼ Rics for s41=2:
Proof. By direct computations, we have for s41=2;
ðKz1; z2Þs ¼XcAZ
ð1þ c2ÞsXqAZ
Tq;cTqþc;�c
ð1þ c2Þs
!/z1ðcÞ; %z2ðcÞS: ð2:13Þ
ARTICLE IN PRESSS. Fang / Journal of Functional Analysis 213 (2004) 440–465452
Let q; cAZ; denote
gðc; qÞ ¼ ð1þ c2Þs � ð1þ q2Þs þ ð1þ ðc� qÞ2Þs: ð2:14Þ
Then we have
�Tc;�qTq;c�q þ Tcþq;�q
ð1þ q2Þs ¼ 1
4
�gðq; cÞgðc; qÞ þ 2gðc; cþ qÞð1þ ðc� qÞ2Þs
ð1þ q2Þsð1þ c2Þsð1þ ðc� qÞ2Þs:
and
Tq;cTqþc;�c
ð1þ c2Þs ¼ 1
4
gðq þ c; qÞ gðq; q þ cÞð1þ q2Þsð1þ c2Þsð1þ ðcþ qÞ2Þs
: ð2:15Þ
By changing q to �q in (2.15), we get
XqAZ
Tq;cTqþc;�c
ð1þ c2Þs ¼ 1
4
XqAZ
gðc; qÞgðq; q � cÞð1þ q2Þsð1þ c2Þsð1þ ðc� qÞ2Þs
:
By (2.14), we see that
�gðq; cÞ � gðq; q � cÞ ¼ �2ð1þ q2Þs;
and
� gðc; qÞð1þ q2Þs þ gðc; cþ qÞð1þ ðc� qÞ2Þs
¼ ð1þ c2Þs½ð1þ ðc� qÞ2Þs � ð1þ q2Þs� þ ½ð1þ q2Þ2s � ð1þ ðc� qÞ2Þsð1þ ðcþ qÞ2Þs�:
It follows that
XqAZ
�Tc;�qTq;c�q þ Tcþq;�q
ð1þ q2Þs �XqAZ
Tq;c Tqþc;�c
ð1þ c2Þs
¼XqAZ
�2gðc; qÞð1þ q2Þs þ 2gðc; cþ qÞð1þ ðc� qÞ2Þs
4ð1þ q2Þsð1þ c2Þsð1þ ðq � cÞ2Þs
¼ 1
2
XqAZ
ð1þ c2Þs½ð1þ ðc� qÞ2Þs � ð1þ q2Þs�ð1þ q2Þsð1þ c2Þsð1þ ðq � cÞ2Þs
þ 1
2
XqAZ
ð1þ q2Þ2s � ð1þ ðc� qÞ2Þsð1þ ðcþ qÞ2Þs
ð1þ q2Þsð1þ c2Þsð1þ ðq � cÞ2Þs
¼ I1ðcÞ þ I2ðcÞ:
ARTICLE IN PRESSS. Fang / Journal of Functional Analysis 213 (2004) 440–465 453
By taking k ¼ c� q in the summation of I1ðcÞ; we get
I1ðcÞ ¼ �I1ðcÞ:
Therefore I1ðcÞ ¼ 0: I2ðcÞ ¼ 0 follows from (2.5). &
3. Boundedness of Ric1=2
The main purpose of this section is to prove
Theorem 3.1. Let z1; z2ACNðS1;GÞ; then
jRic1=2ðz1; z2Þjpð4pÞjz1j1=2jz2j1=2: ð3:1Þ
Proof. Denote AqðcÞ ¼ 4Að1=2Þq ðcÞ: Then
AqðcÞ ¼1þ 2qc� 2
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffið1þ c2Þð1þ q2Þ
qþ 2½
ffiffiffiffiffiffiffiffiffiffiffiffiffi1þ c2
pþ
ffiffiffiffiffiffiffiffiffiffiffiffiffi1þ q2
p�
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi1þ ðcþ qÞ2
q�ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi1þ ðc� qÞ2
qffiffiffiffiffiffiffiffiffiffiffiffiffi1þ c2
p ffiffiffiffiffiffiffiffiffiffiffiffiffi1þ q2
p ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi1þ ðc� qÞ2
q :
By (1.12), we have
Ric1=2ðz1; z2Þ ¼1
4
XcAZ
ffiffiffiffiffiffiffiffiffiffiffiffiffi1þ c2
p XqAZ
AqðcÞ !
/z1ðcÞ; %z2ðcÞS: ð3:2Þ
Using the following lemma, we get the result. &
Lemma 3.2. We have
supc
XqAZ
jAqðcÞjp16p: ð3:3Þ
Proof. By symmetry, it is sufficient to prove that supcX0
PqAZ jAqðcÞjp16p:
Consider firstly the termP
qX0 jAqðcÞj: Define
A1qðcÞ ¼
qc�ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffið1þ c2Þð1þ q2Þ
qffiffiffiffiffiffiffiffiffiffiffiffiffi1þ c2
p ffiffiffiffiffiffiffiffiffiffiffiffiffi1þ q2
p ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi1þ ðc� qÞ2
q ;
A2qðcÞ ¼
ffiffiffiffiffiffiffiffiffiffiffiffiffi1þ c2
pþ
ffiffiffiffiffiffiffiffiffiffiffiffiffi1þ q2
p�
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi1þ ðcþ qÞ2
qffiffiffiffiffiffiffiffiffiffiffiffiffi1þ c2
p ffiffiffiffiffiffiffiffiffiffiffiffiffi1þ q2
p :
ARTICLE IN PRESSS. Fang / Journal of Functional Analysis 213 (2004) 440–465454
We have
qcffiffiffiffiffiffiffiffiffiffiffiffiffi1þ c2
p ffiffiffiffiffiffiffiffiffiffiffiffiffi1þ q2
p � 1 ¼ qffiffiffiffiffiffiffiffiffiffiffiffiffi1þ q2
p � 1
!� 1� cffiffiffiffiffiffiffiffiffiffiffiffiffi
1þ c2p
!qffiffiffiffiffiffiffiffiffiffiffiffiffi
1þ q2p
¼� 1ffiffiffiffiffiffiffiffiffiffiffiffiffi1þ q2
pðq þ
ffiffiffiffiffiffiffiffiffiffiffiffiffi1þ q2
p� 1ffiffiffiffiffiffiffiffiffiffiffiffiffi
1þ c2p
ðcþffiffiffiffiffiffiffiffiffiffiffiffiffi1þ c2
pÞ
qffiffiffiffiffiffiffiffiffiffiffiffiffi1þ q2
p :
Let
B1qðcÞ ¼ � 1ffiffiffiffiffiffiffiffiffiffiffiffiffi
1þ q2p
ðq þffiffiffiffiffiffiffiffiffiffiffiffiffi1þ q2
pÞ
1ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi1þ ðc� qÞ2
q ;
B2qðcÞ ¼
1ffiffiffiffiffiffiffiffiffiffiffiffiffi1þ c2
pðcþ
ffiffiffiffiffiffiffiffiffiffiffiffiffi1þ c2
pÞ
qffiffiffiffiffiffiffiffiffiffiffiffiffi1þ q2
p 1ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi1þ ðq þ cÞ2
q � 1ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi1þ ðq � cÞ2
q0B@
1CA;
B3qðcÞ ¼ � 1ffiffiffiffiffiffiffiffiffiffiffiffiffi
1þ c2p
ðcþffiffiffiffiffiffiffiffiffiffiffiffiffi1þ c2
pÞ
qffiffiffiffiffiffiffiffiffiffiffiffiffi1þ q2
p 1ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi1þ ðq þ cÞ2
q :
Then
A1qðcÞ ¼ B1
qðcÞ þ B2qðcÞ þ B3
qðcÞ: ð3:4Þ
Clearly
jB1qðcÞjp
1
1þ q2and
Xcq¼0
jB2qðcÞjp1: ð3:5Þ
We have
1ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi1þ ðq þ cÞ2
q � 1ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi1þ ðq � cÞ2
q
¼
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi1þ ðq � cÞ2
q�
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi1þ ðq þ cÞ2
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi1þ ðq þ cÞ2
q ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi1þ ðq � cÞ2
q¼ �4qcffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
1þ ðq þ cÞ2q ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
1þ ðq � cÞ2q
ðffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi1þ ðq � cÞ2
qþ
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi1þ ðq þ cÞ2
qÞ:
ARTICLE IN PRESSS. Fang / Journal of Functional Analysis 213 (2004) 440–465 455
It follows that
Xq4c
jB2qðcÞjp
XqX1
1
1þ q2oþN: ð3:6Þ
For the estimate of A2qðcÞ; we have
ffiffiffiffiffiffiffiffiffiffiffiffiffi1þ q2
p�
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi1þ ðcþ qÞ2
qffiffiffiffiffiffiffiffiffiffiffiffiffi1þ c2
p ffiffiffiffiffiffiffiffiffiffiffiffiffi1þ q2
p ¼ �c2 � 2qcffiffiffiffiffiffiffiffiffiffiffiffiffi1þ c2
p ffiffiffiffiffiffiffiffiffiffiffiffiffi1þ q2
pðffiffiffiffiffiffiffiffiffiffiffiffiffi1þ q2
pþ
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi1þ ðcþ qÞ2
qÞ:
Let
C1qðcÞ ¼
1ffiffiffiffiffiffiffiffiffiffiffiffiffi1þ q2
p � cþ 2qffiffiffiffiffiffiffiffiffiffiffiffiffi1þ q2
pðffiffiffiffiffiffiffiffiffiffiffiffiffi1þ q2
pþ
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi1þ ðcþ qÞ2
q ;
C2qðcÞ ¼
cffiffiffiffiffiffiffiffiffiffiffiffiffi1þ c2
pðcþ
ffiffiffiffiffiffiffiffiffiffiffiffiffi1þ c2
pÞ
1ffiffiffiffiffiffiffiffiffiffiffiffiffi1þ q2
pðffiffiffiffiffiffiffiffiffiffiffiffiffi1þ q2
pþ
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi1þ ðcþ qÞ2
qÞ;
C3qðcÞ ¼
1ffiffiffiffiffiffiffiffiffiffiffiffiffi1þ c2
pðcþ
ffiffiffiffiffiffiffiffiffiffiffiffiffi1þ c2
pÞ
qffiffiffiffiffiffiffiffiffiffiffiffiffi1þ q2
p 2ffiffiffiffiffiffiffiffiffiffiffiffiffi1þ q2
pþ
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi1þ ðcþ qÞ2
q :
Then
A2qðcÞ ¼ C1
qðcÞ þ C2qðcÞ þ C3
qðcÞ: ð3:7Þ
Clearly
jC2qðcÞjp
1
1þ q2: ð3:8Þ
We have
C1qðcÞ ¼
1ffiffiffiffiffiffiffiffiffiffiffiffiffi1þ q2
pffiffiffiffiffiffiffiffiffiffiffiffiffi1þ q2
pþ
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi1þ ðcþ qÞ2
q� ðq þ cþ qÞffiffiffiffiffiffiffiffiffiffiffiffiffi
1þ q2p
þffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi1þ ðcþ qÞ2
q
¼ 2ffiffiffiffiffiffiffiffiffiffiffiffiffi1þ q2
p 1þffiffiffiffiffiffiffiffiffiffiffiffiffi1þ q2
p ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi1þ ðcþ qÞ2
q� qðcþ qÞ
ðffiffiffiffiffiffiffiffiffiffiffiffiffi1þ q2
pþ
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi1þ ðcþ qÞ2
qÞð
ffiffiffiffiffiffiffiffiffiffiffiffiffi1þ q2
pþ
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi1þ ðcþ qÞ2
qþ cþ 2qÞ
ARTICLE IN PRESSS. Fang / Journal of Functional Analysis 213 (2004) 440–465456
and
ffiffiffiffiffiffiffiffiffiffiffiffiffi1þ q2
p ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi1þ ðcþ qÞ2
q� qðcþ qÞ ¼ 1þ q2 þ ðcþ qÞ2ffiffiffiffiffiffiffiffiffiffiffiffiffi
1þ q2p ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
1þ ðcþ qÞ2q
þ qðcþ qÞ:
It follows that
jC1qðcÞjp
4
1þ q2: ð3:9Þ
We have
2ffiffiffiffiffiffiffiffiffiffiffiffiffi1þ q2
pþ
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi1þ ðcþ qÞ2
q � 1ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi1þ ðq þ cÞ2
q
¼
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi1þ ðq þ cÞ2
q�
ffiffiffiffiffiffiffiffiffiffiffiffiffi1þ q2
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi1þ ðq þ cÞ2
qðffiffiffiffiffiffiffiffiffiffiffiffiffi1þ q2
pþ
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi1þ ðq þ cÞ2
qÞ
¼ c2 þ 2qcffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi1þ ðq þ cÞ2
qðffiffiffiffiffiffiffiffiffiffiffiffiffi1þ q2
pþ
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi1þ ðq þ cÞ2
qÞ2:
It follows that
jB3qðcÞ þ C3
qðcÞjp1
1þ q2: ð3:10Þ
Combining (3.4)–(3.10) and remarking thatP
qX01ffiffiffiffiffiffiffiffi
1þc2p ffiffiffiffiffiffiffiffi
1þq2p ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
1þðc�qÞ2p is dominated
by 1þP
qX11
1þq2; we obtain that
supcX0
XqX0
jAqðcÞjp9XqX0
1
1þ q2
!:
We shall estimate the termP
qo0 jAqðcÞj: We have for q40
A�qðcÞ ¼1� 2qc� 2
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffið1þ c2Þð1þ q2Þ
qþ 2½
ffiffiffiffiffiffiffiffiffiffiffiffiffi1þ c2
pþ
ffiffiffiffiffiffiffiffiffiffiffiffiffi1þ q2
p�
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi1þ ðc� qÞ2
q�ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi1þ ðcþ qÞ2
q�ffiffiffiffiffiffiffiffiffiffiffiffiffi
1þ c2p ffiffiffiffiffiffiffiffiffiffiffiffiffi
1þ q2p ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
1þ ðcþ qÞ2q :
ARTICLE IN PRESSS. Fang / Journal of Functional Analysis 213 (2004) 440–465 457
Let
D1qðcÞ ¼
�2qcþ ðffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi1þ ðcþ qÞ2
q�
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi1þ ðc� qÞ2
qÞffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi1þ ðcþ qÞ2
qffiffiffiffiffiffiffiffiffiffiffiffiffi1þ c2
p ffiffiffiffiffiffiffiffiffiffiffiffiffi1þ q2
p ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi1þ ðcþ qÞ2
q ;
D2qðcÞ ¼
qc�ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffið1þ c2Þð1þ q2Þ
qffiffiffiffiffiffiffiffiffiffiffiffiffi1þ c2
p ffiffiffiffiffiffiffiffiffiffiffiffiffi1þ q2
p ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi1þ ðcþ qÞ2
q þ
ffiffiffiffiffiffiffiffiffiffiffiffiffi1þ c2
pþ
ffiffiffiffiffiffiffiffiffiffiffiffiffi1þ q2
p�
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi1þ ðcþ qÞ2
qffiffiffiffiffiffiffiffiffiffiffiffiffi1þ c2
p ffiffiffiffiffiffiffiffiffiffiffiffiffi1þ q2
p :
We have
D1qðcÞ ¼
1þ ðc2 þ q2 �ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi1þ ðc� qÞ2
q ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi1þ ðcþ qÞ2
qÞffiffiffiffiffiffiffiffiffiffiffiffiffi
1þ c2p ffiffiffiffiffiffiffiffiffiffiffiffiffi
1þ q2p ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
1þ ðcþ qÞ2q :
As
c2 þ q2 �ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi1þ ðc� qÞ2
q ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi1þ ðcþ qÞ2
q¼ 4c2q2 � 1� 2c2 � 2q2
ðc2 þ q2Þ þffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi1þ ðc� qÞ2
q ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi1þ ðcþ qÞ2
q ;
we get
jD1qðcÞjp
1
1þ q2þ 4c
q2 þ c2: ð3:11Þ
According to the fact
XþN
q¼1
c
q2 þ c2pZ þN
0
c
t2 þ c2dt ¼ p
2;
we get
supcX0
XqX1
jD1qðcÞjp2pþ
XqX1
1
1þ q2: ð3:12Þ
For the estimate of D2qðcÞ; we have
D2qðcÞ ¼ � 1ffiffiffiffiffiffiffiffiffiffiffiffiffi
1þ q2p
ðq þffiffiffiffiffiffiffiffiffiffiffiffiffi1þ q2
pÞ
1ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi1þ ðcþ qÞ2
q þ B3qðcÞ þ A2
qðcÞ:
It follows that
jD2qðcÞjp
1
1þ q2þ jB3
qðcÞ þ C3qðcÞj þ jC1
qðcÞj þ jC2qðcÞj ð3:13Þ
ARTICLE IN PRESSS. Fang / Journal of Functional Analysis 213 (2004) 440–465458
which is smaller than 71þq2
; according to (3.10), (3.9) and (3.8). Combining (3.12) and
(3.13), we get
supcX0
Xqo0
jAqðcÞjp2pþ 9XqX1
1
1þ q2:
Finally, we get
XqAZ
jAqðcÞjp2pþ 9þ 18XqX1
1
1þ q2p16p: &
4. Further discussions on jjRicsjjop
The natural question is now whether
sup1=2psp1
jjRicsjjopoþN: ð4:1Þ
In what follows, we shall give a positive response for restriction on subspaces
generated by any finite subdivision. Let P ¼ fy1o?oyNg be a subdivision of S1:
Denote by GðsÞ ¼ ðGðsÞðyi; yjÞÞ1pi; jpN which is positive definite matrix. Let QðsÞ ¼ðQðsÞ
ij Þ1pi; jpN be the inverse of GðsÞ: Consider on GP ¼ G ? G the metric
jaj2P ¼XN
i; j¼1Q
ðsÞij /ai; ajSG; a ¼ ða1;y; aNÞAGP: ð4:2Þ
To a ¼ ða1;y; aNÞAGP; we associate haAHsðGÞ by
haðyÞ ¼XN
i; j¼1GðsÞðyi; yÞQðsÞ
ij aj: ð4:3Þ
By (0.6), we have jhaj2s ¼ jaj2P:
Proposition 4.1. We have for s412;
ðRics ha; haÞs ¼1
4
Xi; j;k;m
XcAZ
LðsÞc
e�ffiffiffiffiffi�1
pðyi�ykÞ
ð1þ c2Þs
!Q
ðsÞij Q
ðsÞkm/aj; amSG ð4:4Þ
ARTICLE IN PRESSS. Fang / Journal of Functional Analysis 213 (2004) 440–465 459
where
LðsÞc ¼
XcAZ
BðsÞq ðcÞ þ
XcAZ
DðsÞq ðcÞ ðsee ð2:4ÞÞ:
Proof. Using (0.8), we have
haðcÞ ¼XN
i; j¼1
e�ffiffiffiffiffi�1
pcyi
ð1þ c2Þs QðsÞij aj:
Now applying (1.12) and (2.4), we get the result. &
Theorem 4.2. For P given, there exists a constant CP40 (independent of s) such that
jðRics ha; haÞsjpCPjhaj2s ; sA1
2; 1
� �: ð4:5Þ
To prove (4.5), we need some preparations.
Lemma 4.3. Let EðsÞ1 ðcÞ ¼
Pjqjpjcj D
ðsÞq ðcÞ: Then
XcAZ
EðsÞ1 ðcÞ
ð1þ c2Þs þXcAZ
1
ð1þ c2Þs
XqAZ
BðsÞq ðcÞ
����������p2as þ 6p: ð4:6Þ
Proof. For cX1; we have
EðsÞ1 ¼ �2 1� 1
ð1þ c2Þs
� �þ 2
Xcq¼1
DðsÞq ðcÞ: ð4:7Þ
By (2.9), for 1pqpc; we have
2ð1þ c2Þs � 2ð1þ q2Þspð1þ ðcþ qÞ2Þs þ ð1þ ðc� qÞ2Þs � 2ð1þ q2Þsp2c2s:
Then
Xcq¼1
DðsÞq ðcÞ þ 2
Xcq¼1
1
ð1þ q2Þs
����������p 2c
ð1þ c2Þsp2:
ARTICLE IN PRESSS. Fang / Journal of Functional Analysis 213 (2004) 440–465460
According to (4.7), we get
XcX1
EðsÞ1 ðcÞ
ð1þ c2Þs þ 4XNc¼1
Xcq¼1
1
ð1þ q2Þsð1þ c2Þs
����������
p6XcX1
1
ð1þ c2Þsp3p: ð4:8Þ
Remark that a2s ¼ 4P
q;cX11
ð1þq2Þsð1þc2Þs þ 4P
qX11
ð1þq2Þs þ 1: By (4.8), we get
XcAZ
EðsÞ1 ðcÞ
ð1þ c2Þs þ a2s
����������p2as þ 6p:
Now remarking that
XcAZ
1
ð1þ c2Þs
XqAZ
BðsÞq ðcÞ ¼
Xc;qAZ
1
ð1þ c2Þsð1þ q2Þs;
we get (4.6). &
Lemma 4.4. Let EðsÞ2 ðcÞ ¼
Pjqj4jcj D
ðsÞq ðcÞ: Then for 1=2psp1;
jEðsÞ2 ðcÞjp8 2þ
XqX1
1
ð1þ q2Þ3=2
!p8p: ð4:9Þ
Proof. Let c40: By (2.10), we get
Xq4c
jDðsÞq ðcÞjp 4s
Xq4c
c2
ð1þ c2Þsð1þ q2Þsð1þ ðq � cÞ2Þ2�s
þ 4sð2s � 1ÞXq4c
c2
ð1þ c2Þsð1þ q2Þsð1þ ðq � cÞ2Þ1�s
p 4sXqX1
1
ð1þ q2Þ1þsþ 4sð2s � 1Þ
XqX1
1
ð1þ q2Þs:
Remarking that
XqX1
1
ð1þ q2Þsp1þZ
N
1
dx
x2s¼ 2s
2s � 1;
we get (4.9). &
ARTICLE IN PRESSS. Fang / Journal of Functional Analysis 213 (2004) 440–465 461
Lemma 4.5. Let gðsÞc ¼Pc
q¼1 DðsÞq ðcÞ; then
XcX1
jgðsÞcþ1 � gðsÞc jp4sþ1as þ 5p: ð4:10Þ
Proof. Write down
gðsÞcþ1 � gðsÞc ¼ �Xcþ1q¼1
ð1þ ðq þ cþ 1Þ2Þs þ ð1þ ðcþ 1� qÞ2Þs � 2ð1þ q2Þs
ðð1þ ðcþ 1Þ2Þsð1þ ðqÞ2Þs
þXcq¼1
ð1þ ðq þ cÞ2Þs þ ð1þ ðc� qÞ2Þs � 2ð1þ q2Þs
ðð1þ ðcÞ2Þsð1þ ðqÞ2Þs
¼ � ð1þ ð2ðcþ 1ÞÞ2Þs þ 1
ð1þ ðcþ 1Þ2Þ2sþ 2
cþ 1
ð1þ ðcþ 1Þ2Þs� c
ð1þ c2Þs
!
þXcq¼1
1þ ðq þ cÞ2Þs
ð1þ c2Þs � 1þ ðq þ cþ 1Þ2Þs
ð1þ ðcþ 1Þ2Þs
!1
ð1þ q2Þs
þXc�1q¼1
1þ ðc� qÞ2Þs
ð1þ c2Þs � 1þ ðcþ 1� qÞ2Þs
ð1þ ðcþ 1Þ2Þs
!1
ð1þ q2Þs
þ 1
ð1þ c2Þs �2s
ð1þ ðcþ 1Þ2Þs
!1
ð1þ c2Þs
¼ I1ðcÞ þ I2ðcÞ þ I3ðcÞ þ I4ðcÞ þ I5ðcÞ:
We have
(i)XcX1
jI1ðcÞjp4sXcX1
1
ð1þ c2Þs þXcX1
1
ð1þ c2Þ2s:
(ii)XcX1
jI5ðcÞjpXcX1
1þ 2s
ð1þ c2Þ2s:
For estimatingP
cX1 jI2ðcÞj; we remark that the function x- xð1þx2Þs is increasing
over ½1; 1ffiffiffiffiffiffiffiffi2s�1
p �; decreasing over ½ 1ffiffiffiffiffiffiffiffi2s�1
p ;N½: Let Ns ¼ ½ 1ffiffiffiffiffiffiffiffi2s�1
p �: Then
(iii)XcX1
jI2ðcÞjp2Ns
ð1þ N2s Þ
s þNs þ 1
ð1þ ðNs þ 1Þ2Þs
!p4:
ARTICLE IN PRESSS. Fang / Journal of Functional Analysis 213 (2004) 440–465462
ForP
cX1 jI3ðcÞj; we remark that the function x-ð1þðqþxÞ2Þs
ð1þx2Þs is decreasing over
½q;þN½: Therefore
jI3ðcÞj ¼Xcq¼1
1þ ðq þ cÞ2Þs
ð1þ c2Þsð1þ q2Þs �Xcq¼1
1þ ðq þ cþ 1Þ2Þs
ð1þ ðcþ 1Þ2Þsð1þ q2Þs:
It follows that
(iv)XcX1
jI3ðcÞjp2XcX1
ð1þ 4c2Þs
ð1þ c2Þ2sp2 4s
XcX1
1
ð1þ c2Þs:
In the same way, we get
(v)XcX2
jI4ðcÞjp2sXcX1
1
ð1þ c2Þ2sþXcX1
1
ð1þ c2Þs:
Combining (i)–(v), we get the result. &
Proof of Theorem 4.2. For iak; we have
XcAZ
XqAZ
BðsÞq ðcÞ
!e�ffiffiffiffiffi�1
pcðyi�ykÞ
ð1þ c2Þs ¼ ðGðsÞðyi; ykÞÞ2: ð4:11Þ
According to (4.7),
XcAZ
EðsÞ1 ðcÞ e�
ffiffiffiffiffi�1
pcðyi�ykÞ
ð1þ c2Þs ¼ 2XcX1
EðsÞ1 ðcÞ cosðcðyi � ykÞÞ
ð1þ c2Þs
¼ � 4XcX1
1� 1
ð1þ c2Þs
� �cosðcðyi � ykÞÞ
ð1þ c2Þs þ 4XcX1
gccosðcðyi � ykÞÞ
ð1þ c2Þs :
Let GðsÞN ¼
PNc¼1
cosðcðyi�ykÞÞð1þc2Þs ; Vcðyi; ykÞ ¼
PNc¼1 cosðcðyi � ykÞÞ and
aP ¼ supN;i;k
jVNðyi; ykÞj:
By Abel transformation, we get
supNX1
jGðsÞN jpjaPj
2s:
ARTICLE IN PRESSS. Fang / Journal of Functional Analysis 213 (2004) 440–465 463
Again by Abel transformation, we obtain
XcX1
gccosðcðyi � ykÞÞ
ð1þ c2Þs
� �����������p
XcX1
jgc � gcþ1j !
supN
jGðsÞN j
� �:
Now using (4.9) and (4.10), we get
XcAZ
XqAZ
DðsÞq ðcÞ
!e�ffiffiffiffiffi�1
pcðyi�ykÞ
ð1þ c2Þs
����������
pð4þ 8pÞas þ 4ð2sas þ 5pÞaP: ð4:12Þ
Finally combining (4.6), (4.9), (4.11) and (4.12) and by (4.4), there exists a constantAP40 (independent of sA½1=2; 1�) such that
jðRics ha; haÞsjpAP
as
Xi; j;k;m
jasQðsÞij j jasQ
ðsÞk;mj j/aj; amSGj:
As s-1=2; asQðsÞij -dij : Let
BP ¼ supij
sup1=2osp1
jasQðsÞij joþN:
Then
jðRics ha; haÞsjpAPB2
P
as
N3XN
j¼1jaj j2G
!:
Let lPasbe the smallest eigenvalue of QðsÞ; then
jðRics ha; haÞsjpAPB2
PN3
lP
XN
i; j¼1Q
ðsÞij /ai; ajSG
!¼ APB2
PN3
lPjhaj2s : &
Remark 4.6. Let HPs ðGÞ ¼ fha; aAGPg and pP : HsðGÞ-HP
s ðGÞ be the orthogo-nal projection, then pP Rics pP is different of Ricci tensor Rics
P on HPs ðGÞ defined
by the induced metric. For this last one, we know that RicsPB
as
4Identity as s-1
2:
References
[Ca] T.R. Carson, Logarithmic Sobolev inequalities for free loop groups, Ph.D. Thesis, University of
California, San Diego, 1997.
[DL] B.K. Driver, T. Lohrenz, Logarithmic Sobolev inequalities for pinned loop groups, J. Funct. Anal.
140 (1996) 381–448.
ARTICLE IN PRESSS. Fang / Journal of Functional Analysis 213 (2004) 440–465464
[Dr] B.K. Driver, Integration by parts and quasi-invariance for heat kernel measures on loop groups,
J. Funct. Anal. 149 (1997) 470–547.
[Fa] S. Fang, Integration by parts formula and logarithmic Sobolev inequality on the path space over
loop groups, Ann. Probab. 27 (1999) 664–683.
[FF] S. Fang, J. Franchi, De Rham–Hodge–Kodaira operator on loop groups, J. Funct. Anal. 148 (1997)
391–407.
[FG] S. Fang, F. Gong, Metrics Hs on free loop groups, Preprint 2002.
[Fr] S. Freed, The geometry of loop groups, J. Differential Geom. 28 (1988) 223–276.
[In] Y. Inahama, Logarithmic Sobolev inequality on free loop groups for heat kernel measures
associated with the general Sobolev spaces, J. Funct. Anal. 179 (2001) 170–213.
[Ma] P. Malliavin, Hypoellipticity in infinite dimension, in: M. Pinsky (Ed.), Diffusion Processes and
Related Problems in Analysis, Vol. I (Chicago 1989), Birkhauser, Basel, 1991, pp. 17–33.
[PS] A. Pressley, G. Segal, Loop Group, Oxford Mathematical Monographs, Clarendon, Oxford, 1986.
ARTICLE IN PRESSS. Fang / Journal of Functional Analysis 213 (2004) 440–465 465